shithub: libvpx

Info  •  Files  •  Log  •  Branches

Refactor selected partition size coding for rtc

This commit makes a refactoring of the rtc_use_partition. It allows
the encoder to take a preferred block size for non-RD mode decision.
The boundary blocks are handled such that smaller block sizes that
fit in the boundary size will be used instread.

In rtc mode, the coding performance of speed -6 for pedestrian_1080p
goes from
158980 b/f, 38.934 dB, 22721 ms to
159008 b/f, 40.064 dB, 23721 ms.

For rtc set, the speed -6 compression performance is improved by
26%. Still about 2dB behind speed -5 at this point.

Change-Id: If0944f0880eaf1ad340bc325d97cea8d0f9dd53f

--- a/vp9/encoder/vp9_encodeframe.c

+++ b/vp9/encoder/vp9_encodeframe.c

@@ -2261,15 +2261,17 @@

   VP9_COMMON *const cm = &cpi->common;

   MACROBLOCK *const x = &cpi->mb;

   MACROBLOCKD *const xd = &cpi->mb.e_mbd;

+  int mis = cm->mode_info_stride;

+  int br, bc;

   int i, j;

   int chosen_rate = INT_MAX;

   int64_t chosen_dist = INT_MAX;

   MB_PREDICTION_MODE mode = DC_PRED;

-  int row8x8_remaining = MIN(MI_BLOCK_SIZE, tile->mi_row_end - mi_row);

-  int col8x8_remaining = MIN(MI_BLOCK_SIZE, tile->mi_col_end - mi_col);

+  int rows = MIN(MI_BLOCK_SIZE, tile->mi_row_end - mi_row);

+  int cols = MIN(MI_BLOCK_SIZE, tile->mi_col_end - mi_col);

-  int rows = mi_row + row8x8_remaining;

-  int cols = mi_col + col8x8_remaining;

+  int mi_8x8_width = num_8x8_blocks_wide_lookup[bsize];

+  int mi_8x8_hight = num_8x8_blocks_high_lookup[bsize];

   int brate;

   int64_t bdist;

@@ -2277,17 +2279,27 @@

   *dist = 0;

   // find prediction mode for each 8x8 block

-  for (j = mi_row; j < rows; ++j) {

-    for (i = mi_col; i < cols; ++i) {

-      set_offsets(cpi, tile, j, i, BLOCK_8X8);

+  for (br = 0; br < rows; br += mi_8x8_hight) {

+    for (bc = 0; bc < cols; bc += mi_8x8_width) {

+      int row = mi_row + br;

+      int col = mi_col + bc;

+      int bh = 0, bw = 0;

+      BLOCK_SIZE bs = find_partition_size(bsize, rows - br, cols - bc,

+                                          &bh, &bw);

+      set_offsets(cpi, tile, row, col, bs);

+

       if (cm->frame_type != KEY_FRAME)

-        vp9_pick_inter_mode(cpi, x, tile, j, i, &brate, &bdist, BLOCK_8X8);

+        vp9_pick_inter_mode(cpi, x, tile, row, col, &brate, &bdist, bs);

       else

-        set_mode_info(&xd->mi_8x8[0]->mbmi, BLOCK_8X8, mode, j, i);

+        set_mode_info(&xd->mi_8x8[0]->mbmi, bs, mode, row, col);

       *rate += brate;

       *dist += bdist;

+

+      for (j = 0; j < bh; ++j)

+        for (i = 0; i < bw; ++i)

+          xd->mi_8x8[j * mis + i] = xd->mi_8x8[0];

     }

   }

@@ -2316,7 +2328,7 @@

     MODE_INFO **mi_8x8 = cm->mi_grid_visible + idx_str;

     cpi->mb.source_variance = UINT_MAX;

-    rtc_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,

+    rtc_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_16X16,

                      &dummy_rate, &dummy_dist, 1);

   }

 }

--- a/vp9/encoder/vp9_pickmode.c

+++ b/vp9/encoder/vp9_pickmode.c

@@ -272,7 +272,7 @@

   if (best_rd > inter_mode_thresh) {

     struct macroblock_plane *const p = &x->plane[0];

     struct macroblockd_plane *const pd = &xd->plane[0];

-    for (this_mode = DC_PRED; this_mode <= H_PRED; ++this_mode) {

+    for (this_mode = DC_PRED; this_mode <= DC_PRED; ++this_mode) {

       vp9_predict_intra_block(xd, 0, b_width_log2(bsize),

                               mbmi->tx_size, this_mode,

                               &p->src.buf[0], p->src.stride,